While these studies have identified specific deficits in neural, cognitive and behavioral domains in AUD, a distinct combination of characteristic features from multiple domains that can successfully classify individuals with AUD diagnosis from unaffected controls has not yet been done. In recent years, Machine Learning approaches have been commonly used to predict and/or classify various neuropsychiatric disorders and outcomes [31,32,33], including AUD [29,34,35]. Random forest (RF), introduced by Leo Breiman [36], is one of the widely used machine learning methods to classify individuals with a particular diagnosis from unaffected controls [37]. According to Sarica et al. [37], the RF method is more protective against overfitting, adaptive to highly non-linear data, and also credible for the data with outliers. Recently, Zhu et al. [29] applied the RF algorithm to successfully classify AUD subjects from control individuals using fMRI based resting state networks and concluded that machine-learning algorithms can serve as alternative techniques to quantify large-scale network differences across clinical groups and to identify of potential biomarkers for a specific diagnostic category.
